Multi-agent learning for winner determination in combinatorial auctions

Abstract

The winner determination problem (WDP) in combinatorial double auctions suffers from computation complexity. In this paper, we attempt to solve the WDP in combinatorial double auctions based on an agent learning approach. Instead of finding the exact solution, we will set up a fictitious market based on multi-agent system architecture and develop a multi-agent learning algorithm to determine the winning bids in the fictitious market to reduce the computational complexity in solving the WDP in combinatorial double auctions. In the fictitious market, each buyer and each seller is represented by an agent. There is a mediator agent that represents the mediator. The issue is to develop learning algorithms for all the agents in the system to collectively solve the winner determination problem for combinatorial double auctions. In this paper, we adopt a Lagrangian relaxation approach to developing efficient multi-agent learning algorithm for solving the WDP in combinatorial double auctions. Numerical results indicate our agent learning approach is more efficient than the centralized approach.